systematic functional annotation of large-scale biological ... · spatial analysis of functional...

of 28 /28
Anastasia Baryshnikova Lewis-Sigler Fellow Princeton University www.baryshnikova-lab.org Systematic Functional Annotation of Large-Scale Biological Networks

Author: others

Post on 17-Jun-2020

3 views

Category:

Documents


0 download

Embed Size (px)

TRANSCRIPT

  • Anastasia Baryshnikova

    Lewis-Sigler Fellow Princeton University

    www.baryshnikova-lab.org

    Systematic Functional Annotation of Large-Scale Biological Networks

    http://www.baryshnikova-lab.org

  • Networks as maps of biological systems

    Martin Krzywinski, Genome Sciences Center, Vancouver, BC www.hiveplot.net

    Good biological interpretation for: • individual interactions • local structures/modules

    Much worse understanding of: • global organization

    http://www.hiveplot.net

  • Gene

    A

    Gene

    B

    Genetic interaction profile correlation > 0.2

    Saccharomyces cerevisiae

    2,838 nodes 10,016 eges

    Costanzo*, Baryshnikova*, et al., Science, 2010

    Genetic interaction similarity network(year 2010)

  • Epistasis

    WT A B AB

    Genetic interactions

    Genome-scale analysis: • Construct & analyze all 36 000 000 combinations of double mutants in yeast • Synthetic Genetic Array (SGA) by Charlie Boone & team (University of Toronto)

    Synthetic lethality

    A

    X

    Y

    Z

    B

    C

    function

    Double mutants with unexpected phenotypes.

    A

    B

    C

    function

    WT A B AB

    Genetic interaction score:ε = AB – A × B

  • Similarity of genetic interaction profiles

    Negative genetic interaction (e.g., synthetic lethality)Positive genetic interaction (e.g., epistasis)

    High correlation

    High correlation

    Low correlation

    Force-directed network layout

    (Cytoscape)

  • Gene

    A

    Gene

    B

    Genetic interaction profile correlation > 0.2

    Saccharomyces cerevisiae

    2,838 nodes 10,016 eges

    Costanzo*, Baryshnikova*, et al., Science, 2010

    Genetic interaction similarity network(year 2010)

  • Costanzo*, Baryshnikova*, et al., Science, 2010

    Gene

    A

    Gene

    B

    Genetic interaction profile correlation > 0.2

    2,838 nodes 10,016 eges

    Genetic interaction similarity network(year 2010)

  • Essential network2016 network

    3,996 nodes, 28,688 edges Average degree = 14.4

    The Yeast Genetic Interaction Similarity Network (Year 2016)

    Many large networks = Need for an automated method for functional annotation.

    2010 network2,838 nodes, 10,016 edges

    Average degree = 7.1

    Non-essential network

  • Spatial Analysis of Functional Enrichment (SAFE)

    1. Take a node A in the network 2. Find all other nodes B that can be reached from A by traveling no more than d 3. Determine whether or not nodes B are statistically enriched for a functional group (e.g., a GO term) 4. Associate A with the –log10 of the enrichment p-value (normalized to [0,1] range).

    B2 is in A’s neighborhood if: d1 + d2 < d

    B1

    A

    d1

    d2B2

    B4

    B5

    B3

    B6

    B7

    B8

  • Different GO terms show different patterns of enrichment

    84% Sparse/small

    4% Multi-regional

    12% Region-specific

  • Related processes = similar patterns of enrichment

    GO:0006895 Golgi to endosome transport

    GO:0022616 DNA strand elongation

    GO:0006888 ER to Golgi vesicle-mediated transport

    Similar enrichment landscapes

    Different enrichment landscapes

    Combine all enrichment landscapes

    (proportionally to their intensities)

    +

    +

    +

  • The automated functional map of the yeast genetic interaction similarity network

    Every color = a group of GO terms enriched in that region

    ~20 seconds

  • The Automated Functional Map of the Yeast Genetic Interaction Similarity Network

    Every color = a group of GO terms enriched in that region

    Every label = the top 5 most frequent words in the names of the GO terms

  • SAFE is sensitive & robust to biological signal

    • It identifies all manually annotated regions + 3 more.

    • It is robust to numerous sources of variation (independent layout runs, distance metrics, neighborhood radius, annotation errors).

    • It is fast & automated → use multiple independent functional standards to annotate the same network.

    - E.g., yeast: 1000s of phenotypic screens, including chemical genomics.

  • The chemical genomic advantage

    A

    B

    C

    function

    X

    Y

    Z

    growth

    A

    B

    C

    function

    X

    Y

    Z

    growth

    drugDrugs can mimic the phenotypes of their mutated targets.

    Test case:132 chemical-genomics screens for drugs with known modes-of-action Hoepfner et al., Microb. Res., 2014

  • Spatial Analysis of Functional Enrichment (SAFE)

    1. Take a node A in the network 2. Find all other nodes B that can be reached from A by traveling no more than d 3. Determine whether or not nodes B are statistically enriched for a functional group (e.g., a GO term) 4. Associate A with the –log10 of the enrichment p-value (normalized to [0,1] range).

    A

    d1

    d2

    d1 + d2 < d

    quantitative phenotype

    0.4

    –1

    –0.7–0.3

    0.02

    0.2–0.8

    –0.1

    Score = 0.4 + 0.2 + 0.02 – 0.1 – 0.3 – 0.7 – 0.8 – 1 = –2.28

    Is this higher or lower than you would expect by random chance?

    B1 B2

    B4

    B5

    B3

    B6

    B7

    B8

  • SAFE recapitulates the known modes-of-action of chemical compounds

    DoxorubicinDNA intercalator, blocks replication

    Gene Ontology Annotation

  • SAFE recapitulates the known modes-of-action of chemical compounds

    Verrucarin AInhibits protein synthesis and mitochondrial function

    DoxorubicinDNA intercalator, blocks replication

  • SAFE uncovers potentially novel mechanisms of drug activityBortezomib

    a.k.a. Velcade and Ps-341

    • The only FDA-approved proteasome inhibitor. • Used for treating multiple myeloma and mantle cell lymphoma. • Promotes programmed cell death. • Synergistic with HDAC inhibitors.

    Proteasome inactivation = accumulation of misfolded proteins = ER stress?

    ?

  • Could it be a side effect of bortezomib?

    Proteasome

    function

    Vesicle-mediatedtransport

    ?

    growth

    Bortezomib

    Proteasome

    function

    Vesicle-mediatedtransport

    ?

    growth

    Bortezomib

    Proteasome

    function

    Vesicle-mediatedtransport

    ?

    growth

    Mutating drug targets can recapitulate their chemical-genetic interactions.

  • Proteasome mutants recapitulate the effect of bortezomib treatment

    Bortezomib rpt4-145

    Proteasome inhibitor ts mutant in an ATPase of the regulatory subunit of the proteasome

  • Spatial Analysis of Functional Enrichment (SAFE)

    • An automated & systematic method for annotating biological networks.

    • Answers 3 fundamental questions:

    1. Are any regions of the network specifically associated with a given function or phenotype?

    2. Where in the network are these regions localized?

    3. How does their localization compare to that of other functions or phenotypes?

    • Builds a functional map of the network and enables the investigation of inter-process relationships.

    • Allows to integrate multiple functional annotations of the same network and gain insight into the molecular mechanisms of drug response.

  • 2016 Yeast Genetic Interaction NetworkThe essential and non-essential networks are both similar and different.

    Essential NetworkNon-essential Network

  • 2016 Yeast Genetic Interaction NetworkComposite Network: with respect to GO biological process, similar to the 2010 network.

  • 2016 Yeast Genetic Interaction NetworkHuh et al., 2003 — Cellular localization: processes appear to organize into larger modules associated with cellular compartments.

  • 2016 Yeast Genetic Interaction NetworkProtein complexes: processes appear to subdivide into smaller modules associated with sets of protein complexes and/or pathways.

  • SAFE starts revealing the functional structure of protein-protein interactions

    5,699 nodes 78,406 edges

    21 functional domains (GO biological process)

  • Acknowledgements

    University of Minnesota

    Chad Myers

    Benjamin VanderSluis Elizabeth Koch

    Carles Pons

    Myers lab

    University of Toronto

    Michael Costanzo Charlie Boone

    Brenda Andrews

    Boone lab Andrews lab

    Princeton University

    Dmitriy Gorenshteyn

    Undergrads Brianna Richardson

    Rachel Xu

    Treeview team Chris Keil

    Lance Parsons Robert Leach

    YeastPhenome team Kara Dolinski Sven Heinicke Rose Oughtred Christie Chang Jennifer Rust

    Mark Schroeder Fan Kang

    www.baryshnikova-lab.org

    www.yeastphenome.org

    SAFE manuscript on Biorxiv (& in press in Cell Systems): http://biorxiv.org/content/early/2015/11/03/030551

    SAFE code (Matlab) + Mac OS X app on Bitbucket: https://bitbucket.org/abarysh/safe

    SAFE app for Cytoscape: Jason Montojo (in progress)

    http://www.baryshnikova-lab.orghttp://www.yeastphenome.orghttp://biorxiv.org/content/early/2015/11/03/030551https://bitbucket.org/abarysh/safe